Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge pull request #2495 from ilya-lavrenov:tapi_flip

Browse Source
pull/2506/merge
Andrey Pavlenko 11 years ago committed by OpenCV Buildbot
parent d18ebfa8e8 04884ebf45
commit 7193762d2d
4 changed files with 51 additions and 40 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 7
      modules/core/src/copy.cpp
  2. 10
      modules/core/src/matrix.cpp
  3. 55
      modules/core/src/opencl/flip.cl
  4. 19
      modules/core/src/opencl/set_identity.cl

7
modules/core/src/copy.cpp
Unescape View File

@ -482,9 +482,9 @@ enum { FLIP_COLS = 1 << 0, FLIP_ROWS = 1 << 1, FLIP_BOTH = FLIP_ROWS | FLIP_COLS
static bool ocl_flip(InputArray _src, OutputArray _dst, int flipCode )
{
CV_Assert(flipCode >= - 1 && flipCode <= 1);
int type = _src.type(), cn = CV_MAT_CN(type), flipType;
int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type), flipType;
if (cn > 4 || cn == 3)
if (cn > 4)
return false;
const char * kernelName;
@ -506,7 +506,8 @@ static bool ocl_flip(InputArray _src, OutputArray _dst, int flipCode )
}
ocl::Kernel k(kernelName, ocl::core::flip_oclsrc,
format( "-D type=%s", ocl::memopTypeToStr(type)));
format( "-D T=%s -D T1=%s -D cn=%d", ocl::memopTypeToStr(type),
ocl::memopTypeToStr(depth), cn));
if (k.empty())
return false;

10
modules/core/src/matrix.cpp
Unescape View File

@ -2679,17 +2679,17 @@ namespace cv {
static bool ocl_setIdentity( InputOutputArray _m, const Scalar& s )
{
int type = _m.type(), cn = CV_MAT_CN(type);
if (cn == 3)
return false;
int type = _m.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
sctype = CV_MAKE_TYPE(depth, cn == 3 ? 4 : cn);
ocl::Kernel k("setIdentity", ocl::core::set_identity_oclsrc,
format("-D T=%s", ocl::memopTypeToStr(type)));
format("-D T=%s -D T1=%s -D cn=%d -D ST=%s", ocl::memopTypeToStr(type),
ocl::memopTypeToStr(depth), cn, ocl::memopTypeToStr(sctype)));
if (k.empty())
return false;
UMat m = _m.getUMat();
k.args(ocl::KernelArg::WriteOnly(m), ocl::KernelArg::Constant(Mat(1, 1, type, s)));
k.args(ocl::KernelArg::WriteOnly(m), ocl::KernelArg::Constant(Mat(1, 1, sctype, s)));
size_t globalsize[2] = { m.cols, m.rows };
return k.run(2, globalsize, NULL, false);

55
modules/core/src/opencl/flip.cl
Unescape View File

@ -39,10 +39,18 @@
//
//M*/
#define sizeoftype ((int)sizeof(type))
#if cn != 3
#define loadpix(addr) *(__global const T *)(addr)
#define storepix(val, addr) *(__global T *)(addr) = val
#define TSIZE (int)sizeof(T)
#else
#define loadpix(addr) vload3(0, (__global const T1 *)(addr))
#define storepix(val, addr) vstore3(val, 0, (__global T1 *)(addr))
#define TSIZE ((int)sizeof(T1)*3)
#endif
__kernel void arithm_flip_rows(__global const uchar* srcptr, int srcstep, int srcoffset,
__global uchar* dstptr, int dststep, int dstoffset,
__kernel void arithm_flip_rows(__global const uchar * srcptr, int src_step, int src_offset,
__global uchar * dstptr, int dst_step, int dst_offset,
int rows, int cols, int thread_rows, int thread_cols)
{
int x = get_global_id(0);
@ -50,19 +58,16 @@ __kernel void arithm_flip_rows(__global const uchar* srcptr, int srcstep, int sr
if (x < cols && y < thread_rows)
{
__global const type* src0 = (__global const type*)(srcptr + mad24(y, srcstep, mad24(x, sizeoftype, srcoffset)));
__global const type* src1 = (__global const type*)(srcptr + mad24(rows - y - 1, srcstep, mad24(x, sizeoftype, srcoffset)));
T src0 = loadpix(srcptr + mad24(y, src_step, mad24(x, TSIZE, src_offset)));
T src1 = loadpix(srcptr + mad24(rows - y - 1, src_step, mad24(x, TSIZE, src_offset)));
__global type* dst0 = (__global type*)(dstptr + mad24(y, dststep, mad24(x, sizeoftype, dstoffset)));
__global type* dst1 = (__global type*)(dstptr + mad24(rows - y - 1, dststep, mad24(x, sizeoftype, dstoffset)));
dst0[0] = src1[0];
dst1[0] = src0[0];
storepix(src1, dstptr + mad24(y, dst_step, mad24(x, TSIZE, dst_offset)));
storepix(src0, dstptr + mad24(rows - y - 1, dst_step, mad24(x, TSIZE, dst_offset)));
}
}
__kernel void arithm_flip_rows_cols(__global const uchar* srcptr, int srcstep, int srcoffset,
__global uchar* dstptr, int dststep, int dstoffset,
__kernel void arithm_flip_rows_cols(__global const uchar * srcptr, int src_step, int src_offset,
__global uchar * dstptr, int dst_step, int dst_offset,
int rows, int cols, int thread_rows, int thread_cols)
{
int x = get_global_id(0);
@ -71,19 +76,16 @@ __kernel void arithm_flip_rows_cols(__global const uchar* srcptr, int srcstep, i
if (x < cols && y < thread_rows)
{
int x1 = cols - x - 1;
__global const type* src0 = (__global const type*)(srcptr + mad24(y, srcstep, mad24(x, sizeoftype, srcoffset)));
__global const type* src1 = (__global const type*)(srcptr + mad24(rows - y - 1, srcstep, mad24(x1, sizeoftype, srcoffset)));
__global type* dst0 = (__global type*)(dstptr + mad24(rows - y - 1, dststep, mad24(x1, sizeoftype, dstoffset)));
__global type* dst1 = (__global type*)(dstptr + mad24(y, dststep, mad24(x, sizeoftype, dstoffset)));
T src0 = loadpix(srcptr + mad24(y, src_step, mad24(x, TSIZE, src_offset)));
T src1 = loadpix(srcptr + mad24(rows - y - 1, src_step, mad24(x1, TSIZE, src_offset)));
dst0[0] = src0[0];
dst1[0] = src1[0];
storepix(src0, dstptr + mad24(rows - y - 1, dst_step, mad24(x1, TSIZE, dst_offset)));
storepix(src1, dstptr + mad24(y, dst_step, mad24(x, TSIZE, dst_offset)));
}
}
__kernel void arithm_flip_cols(__global const uchar* srcptr, int srcstep, int srcoffset,
__global uchar* dstptr, int dststep, int dstoffset,
__kernel void arithm_flip_cols(__global const uchar * srcptr, int src_step, int src_offset,
__global uchar * dstptr, int dst_step, int dst_offset,
int rows, int cols, int thread_rows, int thread_cols)
{
int x = get_global_id(0);
@ -92,13 +94,10 @@ __kernel void arithm_flip_cols(__global const uchar* srcptr, int srcstep, int sr
if (x < thread_cols && y < rows)
{
int x1 = cols - x - 1;
__global const type* src0 = (__global const type*)(srcptr + mad24(y, srcstep, mad24(x, sizeoftype, srcoffset)));
__global const type* src1 = (__global const type*)(srcptr + mad24(y, srcstep, mad24(x1, sizeoftype, srcoffset)));
__global type* dst0 = (__global type*)(dstptr + mad24(y, dststep, mad24(x1, sizeoftype, dstoffset)));
__global type* dst1 = (__global type*)(dstptr + mad24(y, dststep, mad24(x, sizeoftype, dstoffset)));
T src0 = loadpix(srcptr + mad24(y, src_step, mad24(x, TSIZE, src_offset)));
T src1 = loadpix(srcptr + mad24(y, src_step, mad24(x1, TSIZE, src_offset)));
dst1[0] = src1[0];
dst0[0] = src0[0];
storepix(src0, dstptr + mad24(y, dst_step, mad24(x1, TSIZE, dst_offset)));
storepix(src1, dstptr + mad24(y, dst_step, mad24(x, TSIZE, dst_offset)));
}
}

19
modules/core/src/opencl/set_identity.cl
Unescape View File

@ -43,17 +43,28 @@
//
//M*/
#if cn != 3
#define loadpix(addr) *(__global const T *)(addr)
#define storepix(val, addr) *(__global T *)(addr) = val
#define TSIZE (int)sizeof(T)
#define scalar scalar_
#else
#define loadpix(addr) vload3(0, (__global const T1 *)(addr))
#define storepix(val, addr) vstore3(val, 0, (__global T1 *)(addr))
#define TSIZE ((int)sizeof(T1)*3)
#define scalar (T)(scalar_.x, scalar_.y, scalar_.z)
#endif
__kernel void setIdentity(__global uchar * srcptr, int src_step, int src_offset, int rows, int cols,
T scalar)
ST scalar_)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src_index = mad24(y, src_step, mad24(x, (int)sizeof(T), src_offset));
__global T * src = (__global T *)(srcptr + src_index);
int src_index = mad24(y, src_step, mad24(x, TSIZE, src_offset));
src[0] = x == y ? scalar : (T)(0);
storepix(x == y ? scalar : (T)(0), srcptr + src_index);
}
}

Write Preview
Loading…
Cancel
Save